KnoxAI - A Framework for Combinatorial Design Prediction

knoxAI is a modular deep learning framework for graph-based and rule-based machine learning, built with FastAPI. It supports regression, classification, and ranking tasks on graph-structured data, and provides a REST API for model training, evaluation, and prediction. Runs are saved and managed through MLflow.

Quickstart

Docker Instructions

1. Install Docker Desktop
2. Clone this repository
3. run docker build -t knoxai . in the root of this repo
4. run docker run -p 8000:8000 knoxai
5. Open FastAPI Docs at http://localhost:8000/docs

Non-Docker Instructions

1. Clone this repository
2. Install dependencies pip install -r requirements.txt
3. Run uvicorn app.main:api --reload --host 127.0.0.1 --port 8000 in the root directory of this repo
4. For MLflow user-interface run mlflow ui --port 5000 --workers 1 in the root directory of this repo
5. Open FastAPI Docs at http://localhost:7000/docs

Use with Knox or Standalone

Manuscripts

Nicholas Roehner, James Roberts, Andrei Lapets, Dany Gould, Vidya Akavoor, Lucy Qin, D. Benjamin Gordon, Christopher Voigt, and Douglas Densmore. GOLDBAR: A Framework for Combinatorial Biological Design. ACS Synthetic Biology Article ASAP (2024). DOI: https://pubs.acs.org/doi/full/10.1021/acssynbio.4c00296

Webpage

https://www.cidarlab.org/knox